“The best engineers don’t memorize answers. They think.”
Hire with conversations,not questionnaires.
An intelligent AI interviewer that conducts realistic technical interviews, adapts naturally to every response, asks meaningful follow-up questions, evaluates true engineering ability, and generates actionable feedback.
Three inputs, one engine, four things it produces.
Control flow is deterministic and only the language is generated. The engine owns the plan, the coverage counters and the ending condition — the model writes the prose.
Attempts, skips, commit days, role and years — all of it read first.
31 days indexed by module, objectives, tools and day archetype.
Full transcript plus every score, carried through the whole session.
Reasoning engine
Decides which curriculum day is on the table, whether the last answer earned a follow-up, and when the interview has met its coverage floor.
Writes from the day's own objectives, never from a static bank.
Reads the answer for specifics, hedging and real decisions made.
0–3 per answer. Drives the follow-up decision and the difficulty ladder.
Summary, strengths, gaps and next actions — quoting their own words.
Watch it listen, then watch it push back.
Concepts light up as they are said. The follow-up is chosen from what was missing — not from a list written before the candidate opened their mouth.
Explain how transformers use self-attention.
Thirty-one days, lit by how the cohort did.
Every planet is a real curriculum day. Brightness is the cohort's actual record on it — and the interviewer draws its questions from the days a candidate genuinely completed.
Hover a planet to read its objectives. Click to fly to it.
One answer. Four different next questions.
The difficulty ladder moves on every turn. Where it moves is decided by how the last answer scored — not by a script written in advance.
“I used Chroma for the vector store and chunked the plan documents before embedding them.”
“Let me narrow that — on day 9 you loaded embeddings into Chroma. What did you actually configure?”
“You mentioned metadata filtering. What were you filtering on, and why that field?”
“Good. Now defend that chunk size against a reviewer who wants it halved.”
“If retrieval and the agent loop ran as one system under load, where does it break first?”
Pick someone and watch it interview them.
67 hard-won topics and 25 skipped missions across the roster — every one of them a question this agent already knows to ask.